Detection and Classification of Human Action Events from Captured Video Streams

Abstract

Human detection and recognizing their actions from the captured video streams is more difficult and challenging task in the field of image processing. The human action recognition is more complex due to variability in shapes and articulation of human body, motions in the background scene, lighting conditions and occlusion. Human actions are recognized by tracking the selected object over the consecutive frames of gray scale image sequences, initially the background motion of the input video stream is subtracted, and its binary images are constructed, the object which needs to be monitored is selected by enclosing the required pixels within bounding rectangle, by using spatiotemporal interest points (Mo-SIFT). The selected foreground pixels within the bounding rectangle are then trackedusing edge tracking algorithm over the consecutive frames of gray scale images. The features like horizontal stride (HS) and vertical distance (VD) are extracted while tracking and the values of these features from the current frame are subtracted with the previous frame values to know the motion. The obtained results after subtraction are then applied to K-Nearest Neiberhood method to recognizing human action using linear prediction technique. This methodology finds an application where monitoring the human actions is required such as shop surveillance, city surveillance, airports surveillance and other places where security is the prime factor.